Introduction

The efficacy of adjuvant chemotherapy (AC) against resected colorectal liver metastases (CRLM) has been investigated in several randomized controlled trials (RCT), and recent RCT have focused on systemic chemotherapy rather than hepatic arterial infusion therapy.1,2,3,4,5,6,7,8,9,10 The systemic administration of fluorouracil and folinic acid (5FU/LV) for 6 months was investigated in 173 patients with resected CRLM in the FFCD ACHBTH AURC 9002 trial. The recurrence-free survival (RFS) of the AC group was significantly better than that of the surgery alone (SA) group (p = 0.028). However, the overall survival (OS) of the AC group was not significantly better than that of the SA group (p = 0.13).1 Hasegawa K et al. also investigated the efficacy of AC (6 months of systemic oral uracil-tegafur combined with leucovorin [UFT/LV]) in 177 patients. A survival benefit was detected in terms of RFS (p = 0.003), but not OS (p = 0.409).2 Neither of these trials found that AC exhibited significantly greater efficacy than SA against resectable CRLM in terms of OS.

Although the efficacy of and indications for AC for CRLM remain unclear, the administration of AC after CRLM have been resected is already widespread in daily practice, not only in Japan, but also worldwide, as AC has been demonstrated to be effective against stage III colorectal cancer.11,12 Therefore, both of the abovementioned RCT suffered from low patient accrual rates.1,2,13,14 On the other hand, real-world data is increasingly being used to assess clinical effectiveness in daily practice because it is considered to represent the clinical setting better than data obtained in clinical trials.15 Previous studies have suggested that propensity score (PS) matching analysis can be used to perform similar analyses to RCT.16,17 Since it would be difficult to conduct a new RCT in which an SA group was used as the control arm,1,2,13,14 we tried to investigate the effectiveness of and indications for AC for resected CRLM via PS-matching analysis of a large nationwide database.

The purpose of this study was to clarify the effectiveness of and indications for AC for resected CRLM using data collected during a Japanese nationwide survey. In order to obtain results that would be readily applicable to daily practice, we analyzed the data after dividing the patients into synchronous CRLM and metachronous CRLM groups.

Methods

Study Design and Data Sources

The Joint Committee for Nationwide Survey on Colorectal Liver Metastasis is composed of colorectal and hepatic surgeons, medical oncologists, and bio-statisticians, all of whom work at specialized centers in Japan. A nationwide database that contained data regarding CRLM was created by the committee and was made available to the participating institutions of the Japanese Society of Hepato-Biliary-Pancreatic Surgery and the Japanese Society for Cancer of the Colon and Rectum. A nationwide survey of CRLM was conducted in 2014, and data regarding patients that were newly diagnosed with CRLM between 2005 and 2007 were retrospectively registered.18 Modern chemotherapy regimens against colorectal cancer, such as fluorouracil plus leucovorin with oxaliplatin (FOLFOX) and fluorouracil plus leucovorin with irinotecan (FOLFIRI), were approved for clinical use in Japan in 2005. Therefore, we decided to collect the data of patients that were newly diagnosed with CRLM between 2005 and 2007. Patients with resected CRLM that met the following criteria were identified: (1) No history of extrahepatic metastasis, (2) no history of preoperative chemotherapy for CRLM, and (3) underwent R0 resection. There were no limitations with regard to the number of CRLM nodules, and all technically resectable CRLM were included. Patients for whom complete datasets, i.e., data regarding age, sex, the number of CRLM nodules, the largest diameter of the CRLM nodules, the date of diagnosis of CRLM, the date of hepatectomy for CRLM, the timing of the CRLM (synchronous or metachronous), the disease-free interval (DFI) between the resection of the primary tumor and the diagnosis of CRLM, the administration of AC after CRLM resection, survival status, recurrence status, and the date of the last follow-up, were available were included. Cases involving in-hospital mortality were excluded. AC was defined as post-hepatectomy chemotherapy (any regimen, duration, or dose). Then, the overall cohort, i.e., the patients with liver-limited CRLM who underwent R0 resection without preoperative chemotherapy, was divided into synchronous and metachronous CRLM cohorts. Synchronous CRLM was defined as CRLM that was already present at the time of the diagnosis of primary colorectal cancer. PS-matching analysis was performed in each cohort. As for the surveillance schedule, the Japanese Society for Cancer of the Colon and Rectum guidelines recommend performing serial measurements of carcinoembryonic antigen (CEA) and cancer antigen 19-9 (CA19-9) levels every 3 months and thoracoabdominal computed tomography scans every 6 months after the resection of stage I to III colorectal cancer.11 The same schedule or an even more intensive schedule is recommended after the resection of stage IV colorectal cancer or recurrent metastases.

All patient data were collected using an anonymous form. This study was approved by the review boards of the participating institutions.

Propensity Score Matching Analysis

In each cohort, Fisher’s exact test was used to examine the correlations between AC and factors related to the patients’ clinical backgrounds, the original tumor, CRLM, or hepatectomy. Logistic regression analysis was conducted to calculate the PS for receiving AC using factors that exhibited p values of < 0.20 in Fisher’s exact test and for which data were missing in < 25% of all cases. Patients who were treated with AC were matched 1:1 with those that were treated with SA based on their PS, using the optimal matching method.

Statistical Analysis

In each PS-matched cohort, RFS and OS after the initial hepatectomy were compared between the two groups. The patients were also divided into several subgroups based on various preoperative factors, whose prognostic impact had previously been investigated,19,20,21,22,23,24 and the effectiveness of AC was examined in each subgroup. In the metachronous cohort, we divided the patients into early and late metachronous groups according to the DFI from the date of the resection of the primary tumor. The effectiveness of AC was analyzed in each group using DFI cut-off values of 3 months, 6 months, 9 months, 1 year, 1.5 years, and 2 years. Then, the most appropriate DFI cut-off value, i.e., that at which the effectiveness of AC was greatest in the early metachronous group, was determined.

A survival analysis was carried out using the log-rank test and Cox proportional hazards regression analysis. p values of < 0.05 were considered to be statistically significant. All statistical analyses were conducted using the software EZR.25

Results

Propensity Score Matching Analysis

Out of 3820 patients that were registered during the nationwide survey, hepatectomy was performed in 2225 patients. There were 1145, 498, and 647 patients in the overall cohort, synchronous cohort, and metachronous cohort, respectively (Fig. 1). The overall cohort comprised 771 (67.3%) patients that were treated with AC and 374 (32.7%) that were treated with SA. The factors that were used to create the PS are indicated by asterisks in Tables 1, 2, and 3 (see Table 1 for the overall cohort, Table 2 for the synchronous cohort, and Table 3 for the metachronous cohort). The concordance index of the PS for receiving AC was 0.715 (95% confidence interval [95%CI] 0.672–0.763) in the overall cohort, 0.655 (95%CI 0.590–0.721) in the synchronous cohort, and 0.722 (95%CI 0.670–0.774) in the metachronous cohort. After the PS matching, the demographics of the AC and SA groups were well balanced in each cohort. There were 422, 170, and 294 PS-matched patients in the overall, synchronous, and metachronous cohorts, respectively. Thereafter, the subjects of the present investigation were limited to the PS-matched patients.

Fig. 1
figure 1

Flow diagram of the present study

Full size image
Table 1 Patient backgrounds in overall cohort
Full size table
Table 2 Patient backgrounds in synchronous cohort
Full size table
Table 3 Patient backgrounds in metachronous cohort
Full size table

Adjuvant Chemotherapy Regimens

The following AC regimens were used to treat the patients in the PS-matched overall cohort: FOLFOX, 54 cases; FOLFIRI, 3 cases; capecitabine with oxaliplatin (CapeOx), 1 case; 5FU/LV, 21 cases; UFT/LV, 86 cases; S-1, 26 cases; hepatic arterial infusion, 14 cases; doxifluridine, 2 cases; unknown, 4 cases. The duration of the AC was ≥ 3 months in 189 patients (89.6%), while it was < 3 months in 22 patients (10.4%).

Survival Analysis

The median duration of the follow-up period was 79.4 months in the overall cohort. Among the 422 PS-matched patients in the overall cohort, recurrence was identified in 255 cases (60.4%). The AC group exhibited significantly better RFS than the SA group (p = 0.045) (Fig. 2). The associated hazard ratio (HR) was 0.784 (95%CI 0.618–0.995). The median duration of the RFS period was 25.2 months [95%CI 18.6–38.9] in the AC group and 16.4 months [12.2–26.9] in the SA group. The 3- and 5-year RFS rates of the AC group were 45.1% [95%CI 38.3–51.7%] and 40.1% [33.4–46.7%], respectively, and those of the SA group were 39.3% [95%CI 32.6–46.0%] and 36.6% [30.0–43.3%], respectively. The sites of recurrence (intrahepatic alone vs. extrahepatic alone vs. both intra- and extrahepatic) did not differ between the two groups (AC group 51.2 vs. 35.0 vs. 13.0%, SA group 41.8 vs. 37.2 vs. 17.1%, p = 0.393). The frequency of re-resection for recurrence did not differ significantly between the AC and SA groups (43.9 vs. 32.6%, p = 0.070).

Fig. 2
figure 2

Survival curves of the propensity score-matched overall cohort. a The adjuvant chemotherapy (AC) group exhibited significantly better recurrence-free survival (RFS) than the surgery alone (SA) group (p = 0.045). b The AC group also displayed significantly better overall survival (OS) than the SA group (p = 0.027). The 5-year OS rate of the AC group was 66.8%, and that of the SA group was 59.6%

Full size image

The AC group also displayed significantly better OS than the SA group (p = 0.027) (Fig. 2). The associated HR was 0.716 (95%CI 0.532–0.964). The median duration of the OS period was 104.1 months [95%CI 88.4–not available] in the AC group and 86.7 months [62.2–not available] in the SA group. The 5-year OS rate of the AC group was 66.8% [95%CI 59.7–72.9%], and that of the SA group was 59.6% [52.1–66.2%]. Among the patients that underwent re-resection for recurrence, OS after the initial hepatectomy was better in the AC group than in the SA group, but the difference was not statistically significant (median duration 102.6 vs. 66.8 months, p = 0.091).

In the PS-matched synchronous cohort, the AC group also demonstrated significantly better RFS (HR 0.677, 95%CI 0.479–0.956, p = 0.027) and OS (HR 0.642, 95%CI 0.424–0.972, p = 0.036) than the SA group (Fig. 3). However, in the metachronous cohort, the AC group did not exhibit significantly better survival than the SA group in terms of either RFS (HR 0.875, 95%CI 0.651–1.176, p = 0.378) or OS (HR 0.881, 95%CI 0.611–1.270, p = 0.496) (Fig. 4).

Fig. 3
figure 3

Survival curves of the propensity score-matched synchronous cohort. a The AC group demonstrated significantly better RFS than the SA group (p = 0.026). b The AC group exhibited significantly better OS than the SA group (p = 0.035)

Full size image
Fig. 4
figure 4

Survival curves of the propensity score-matched metachronous cohort. a The AC group did not display significantly better RFS than the SA group (p = 0.377). b The AC group did not demonstrate significantly better OS than the SA group (p = 0.496)

Full size image

Subgroup Analyses

In the PS-matched overall cohort, the AC group displayed significantly better RFS and OS than the SA group in the subgroups involving lymph node metastasis from the primary colorectal tumor, synchronous CRLM, a serum CEA level of ≥ 10 ng/ml, a serum CA19-9 level of ≥ 37 IU/ml, a clinical risk score (CRS)19 of ≥ 3, or a CRLM size of < 50 mm (Table 4). The AC group also demonstrated significantly better RFS than the SA group in the subgroup with ≥ 5 CRLM and a CRS of 2 (Supplemental Figs. 1, 2, and 3). Regarding the AC regimens, the patients given oral AC regimens achieved significantly better RFS than those that underwent SA (HR 0.701, 95%CI 0.523–0.939, p = 0.017). As for OS, the patients treated with oral AC regimens also had better prognoses than those treated with SA, but the difference was not significant (HR 0.701, 95%CI 0.490–1.004, p = 0.053). The prognosis of the patients given FOLFOX/CapeOx and that of the patients treated with SA was comparable (RFS HR 1.005, 95%CI 0.704–1.436, p = 0.976; OS HR 0.806, 95%CI 0.502–1.295, p = 0.373).

Table 4 Subgroups analysis of overall cohort
Full size table

In the PS-matched synchronous cohort, AC had beneficial effects on one or both of RFS and OS in the following subgroups: the patients with lymph node metastasis, ≥ 5 CRLM, or a CRLM size of < 50 mm. Also, AC was marginally effective (p < 0.100) in terms of RFS and/or OS in the subgroups involving 2–4 metastases, a serum CEA level of ≥ 10 ng/ml, or a serum CA19-9 level of ≥ 37 IU/ml, all of which displayed HR of < 0.700 (Table 5).

Table 5 Subgroup analysis of synchronous cohort
Full size table

In the PS-matched metachronous cohort, the most appropriate DFI was found to be 1 year. In the early metachronous group (DFI of ≤ 1 year), OS was significantly better in the AC group than in the SA group (OS HR 0.572, 95%CI 0.335–0.979, p = 0.042) (Table 6, Fig. 5). RFS was also better in the AC group, but the difference was not statistically significant (RFS HR 0.667, 95%CI 0.432–1.031, p = 0.068). AC was even more effective than SA in terms of both RFS and OS among the patients with early metachronous CRLM and serum CEA levels of ≥ 10 ng/ml (RFS HR 0.532, 95%CI 0.305–0.929, p = 0.026; OS HR 0.393, 95%CI 0.191–0.807, p = 0.011).

Table 6 Subgroup analysis of metachronous cohort
Full size table
Fig. 5
figure 5

Survival curves of patients with DFI ≤ 1 year in the propensity score-matched metachronous cohort. a The AC group demonstrated marginally better RFS than the SA group (p = 0.066). b The AC group exhibited significantly better OS than the SA group (p = 0.039)

Full size image

Discussion

Although several RCT have demonstrated that administering AC for resected CRLM was effective at increasing RFS, no survival benefit of AC in terms of OS has been demonstrated.1,2,3,4,5,6,7,8,9,10 On the contrary, in the present study, AC was clearly demonstrated to improve both RFS and OS. Regarding the patients’ backgrounds, the present study included more cases of highly malignant CRLM than previous RCT, i.e., more cases involving ≥ 5 CRLM, lymph node metastasis from the primary colorectal tumor, synchronous CRLM, early metachronous CRLM with a DFI of ≤ 1 year, or an elevated CEA level at hepatectomy (Table 7). It was presumed that the current study included more cases with higher CRS. As the effectiveness of AC gradually rose as the CRS increased, the abovementioned differences were considered to be one of the reasons why AC was demonstrated to be effective at promoting OS in this study. Recently, the indications for hepatectomy for CRLM have expanded to include highly malignant disease due to advances in chemotherapy and surgery.12,26,27,28 The patients’ backgrounds of the present study reflected those of the patients that currently undergo surgery for CRLM, and the present study indicated that AC was effective in this clinical settings. Also, the sample size was 422 in the PS-matched overall cohort, and the median duration of the follow-up period was 79.4 months, which were sufficient for investigating the effects of AC on OS. These characteristics of the present study help to explain the significant difference in OS detected between the AC and SA groups.

Table 7 Comparison of the current study with the previous clinical trials
Full size table

Several studies have investigated the factors associated with prognosis after hepatectomy for CRLM, such as lymph node metastasis, a DFI of ≤ 1 year after the resection of the primary colorectal tumor, multiple CRLM, a CRLM size of > 50 mm, synchronous extrahepatic metastases, a serum CEA level of > 200 ng/ml, and a serum CA19-9 level of > 100 U/ml.19,20,21,22,23,24 In the PS-matched overall cohort, AC had beneficial effects on one or both of RFS and OS in various subgroups, such as those involving lymph node metastasis, synchronous CRLM, ≥ 5 CRLM, a serum CEA level of ≥ 10 ng/ml, a serum CA19-9 level of ≥ 37 IU/ml, or a CRS of ≥ 2. These highly malignant cases of CRLM clearly benefited from AC and met the indications for AC. Regarding the size of CRLM, AC did not demonstrate a beneficial effect in the subgroup of patients with larger tumors (≥ 50 mm) in this study. In our previous study,20 a tumor size of > 50 mm was found to have a minimal impact on RFS and did not have a significant impact on OS. Therefore, large CRLM do not appear to be highly malignant. This fact, as well as the small number of patients with large tumors, might explain why AC was not shown to be effective in the ≥ 50 mm CRLM subgroup.

We also investigated the survival benefit of and indications for AC in synchronous and metachronous CRLM cohorts. In the PS-matched synchronous CRLM cohort, AC was effective in terms of both RFS and OS. AC had beneficial effects on RFS and/or OS in various subgroups, such as those involving lymph node metastasis, ≥ 5 CRLM, or a CRLM size of < 50 mm. Also, AC was effective in terms of RFS and/or OS in the subgroups involving 2–4 metastases, a serum CEA level of ≥ 10 ng/ml, or a serum CA19-9 level of ≥ 37 IU/ml, although these differences were not significant (p < 0.100). As AC has demonstrated clear survival benefits when used against stage III colorectal cancer, synchronous CRLM should be considered to be a favorable indication for AC, especially in cases involving lymph node metastasis, ≥ 2 CRLM, a serum CEA level of ≥ 10 ng/ml, or a serum CA19-9 level of ≥ 37 IU/ml.

On the other hand, in the PS-matched metachronous CRLM cohort, AC was not markedly effective in terms of RFS or OS. However, in the subgroup involving DFI of ≤ 1 year, AC was significantly effective in terms of OS. Also, its significance was even more evident among the patients that exhibited both serum CEA levels of ≥ 10 ng/ml and DFI of ≤ 1 year. Previous studies have indicated that a DFI of ≤ 1 year is a risk factor for recurrence and should be used as an indication for AC.19,29,30 Our findings are compatible with these previous results. On the contrary, in the subgroups involving DFI of > 1 year, AC was not effective in terms of both RFS and OS. Neoadjuvant chemotherapy for CRLM in patients that were with low risk of recurrence was reported to provide no survival benefit compared with upfront surgery,31 which was similar to the present study. Therefore, we consider that early metachronous CRLM with a DFI of ≤ 1 year is a good indication for AC, whereas AC should be cautiously indicated for late metachronous CRLM with a DFI of > 1 year, after taking account of other clinical risk factors.

Regarding AC regimens, the present study included various regimens, such as FOLFOX/CapeOx (26.1%), 5FU/LV (10.0%), UFT/LV (40.8%), and S-1 (12.3%). As FOLFOX was only approved for clinical use in Japan in 2005, the proportion of patients treated with FOLFOX/CapeOx was relatively small. These characteristics represent the situation encountered in current daily practice in Japan, and so the findings of the present study indicate the effectiveness of AC in this clinical setting. The patients given oral regimens demonstrated significantly better RFS than those treated with SA. As FOLFOX/CapeOx tended to be offered for more advanced disease, it was not possible to assess whether these regimens were better than others, but oral regimens might be effective against CRLM, as demonstrated by a previous RCT of UFT/LV.2 Taking account of the high recurrence rate and peripheral sensory neuropathy seen after oxaliplatin treatment, it might be worth considering a strategy involving the administration of oral AC regimens followed by treatment with FOLFOX/CapeOx plus targeted drugs for recurrence.

The limitations of the present study include the fact that it was not an RCT. However, it would be difficult to conduct a new RCT in which an SA group was used as the control arm.1,2,13,14 As PS-matching studies based on large databases are considered to be an alternative to RCT,16,17 we consider that the current study indicates the effectiveness of AC in the clinical setting. Second, the current study was based on a nationwide survey of 134 institutions in Japan, and the treatment strategies of the non-academic centers might have influenced the results. However, the nationwide survey reflected the current status of treatment for CRLM, and so the present study indicates the effectiveness of AC in clinical practice. Third, it was not possible to investigate in detail the effects of treatments that were administered for post-hepatectomy recurrence. OS after the initial resection of CRLM is reported to be influenced by the treatment employed for recurrence.1,2,3,32 However, because the median duration of the OS period was longer in the AC group than in the SA group, even among patients who underwent re-resection, we consider that AC is effective against CRLM.

Conclusion

Adjuvant chemotherapy for resected CRLM had a survival benefit in terms of both RFS and OS in the clinical setting. Synchronous CRLM is a favorable indication for AC, whereas in metachronous CRLM the use of AC should be individualized according to each patient’s risk factors.